Joint Latency and Energy Consumption for Federated Learning Over NOMA-UAV Networks

Federated learning (FL) attracts widespread attention due to its powerful resource aggregation capabilities and advantages in privacy protection. However, during the model training process, restricted by limited communication resources and poor channel conditions, it is challenging to achieve a large number of devices access while ensuring signal strength, which becomes a bottleneck affecting the performance of FL. In this paper, to address this issue, we propose a FL framework based on non-orthogonal multiple access (NOMA) and utilize unmanned aerial vehicles (UAVs) to assist in establishing line-of-sight channels. We jointly optimize the latency and energy consumption of the proposed FL framework considering the impact of device selection imposed on the model convergence. Finally, we solve the optimization problem using the proximal policy optimization algorithm and verify its superiority.